Research map / Vision-language and multimodal / Robot action and navigation
RoboStream: Weaving Spatio-Temporal Reasoning with Memory in Vision-Language Models for Robotics
Yuzhi Huang, Jie Wu, Weijue Bu, Ziyi Xiong, Gaoyang Jiang, Ye Li, Kangye Ji, Shuzhao Xie, Yue Huang, Chenglei Wu, Jingyan Jiang, Zhi Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add CSTG (ablation) | RoboStream-235B · RLBench long-horizon average | success rate (%): 14.5 -> 90.5 | Improved |
| add STF-Tokens on top of CSTG (ablation) | RoboStream-235B · RLBench long-horizon average | success rate (%): 79.5 -> 90.5 | Improved |
| add STF-Tokens without CSTG (ablation) | RoboStream-235B · RLBench long-horizon average | success rate (%): 12.0 -> 14.5 | Improved |
| add CSTG (ablation) | RoboStream-8B · RLBench long-horizon average | success rate (%): 12.5 -> 58.0 | Improved |
| add CSTG (ablation) | RoboStream-32B · RLBench long-horizon average | success rate (%): 8.5 -> 82.5 | Improved |
| add STF-Tokens on top of CSTG (ablation) | RoboStream-8B · RLBench long-horizon average | success rate (%): 45.5 -> 58.0 | Improved |
| add STF-Tokens on top of CSTG (ablation) | RoboStream-32B · RLBench long-horizon average | success rate (%): 69.0 -> 82.5 | Improved |
| scale VLM backbone from 8B to 235B | RoboStream · RLBench long-horizon average | success rate (%): 58.0 -> 90.5 | Improved |
| use RoboStream-8B instead of SoFar | SoFar · RLBench long-horizon average | success rate (%): 28.0 -> 58.0 | Improved |
| use RoboStream-8B instead of SoFar (zero-shot) | SoFar · SIMPLER WidowX+Bridge average | success rate (%): 58.3 -> 74.8 | Improved |
| use RoboStream-32B instead of SoFar-32B | SoFar-32B · 6-DoF SpatialBench total | accuracy (%): 45.3 -> 48.9 | Improved |
| use RoboStream-32B instead of SoFar-32B | SoFar-32B · Open6DOR V2 6-DoF | success rate (%): 48.4 -> 52.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SoFar — SoFar: Language-Grounded Orientation Bridges Spatial Reasoning and Object Manipulation (NeurIPS 2025) · better on 12 of 12 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- ConsisVLA-4D: Advancing Spatiotemporal Consistency in Efficient 3D-Perception and 4D-Reasoning for Robotic Manipulation (CVPR 2026 · similarity 0.92)
- Seeing Across Views: Benchmarking Spatial Reasoning of Vision-Language Models in Robotic Scenes (ICLR 2026 · similarity 0.89)
- Linear Mechanisms for Spatiotemporal Reasoning in Vision Language Models (ICLR 2026 · similarity 0.89)
- RoboSpatial: Teaching Spatial Understanding to 2D and 3D Vision-Language Models for Robotics (CVPR 2025 · similarity 0.89)
- RoboRefer: Towards Spatial Referring with Reasoning in Vision-Language Models for Robotics (NeurIPS 2025 · similarity 0.89)
- CausalSpatial: A Benchmark for Object-Centric Causal Spatial Reasoning (NeurIPS 2026 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
